Font Size: a A A

Research On Light Path Reuse Techniques In Ray Tracing Rendering

Posted on:2017-05-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:H QinFull Text:PDF
GTID:1318330518973529Subject:Computer Science and Technology
Abstract/Summary:
Photorealistic rendering is one of the most important problems in computer graphics.In pho-torealistic rendering,Monte Carlo ray tracing is one of the most general methods.The method can handle arbitrary lighting,scene geometry,texture or camera settings.However,its huge computa-tional complexity also makes it highly time consuming.Path reuse is an important way to reduce the render time of Monte Carlo methods.By reusing path components,a single path can make multiple contributions to the light transport estimation,which reduces the amortized sampling cost.Additionally,reusing paths based on their importance can serve as a form of importance sampling.Path reuse can be divided into three categories based on the how the reused path samples are created.In bidirectional path reuse,we reuse a sensor sub-path and connect it with multiple light sub-paths to create complete path samples,or vice versa.In unidirectional path reuse,we reuse the starting part of one path sample and connect it with the finishing parts of other path samples in the same sampling direction.In image space shading reuse,we directly reuse the estimation results of path samples in the 2D image space.This is also equivalent to reusing the whole path sample.This thesis systematically studies the path reuse techniques in Monte Carlo ray tracing.Our three major works have covered all three categories of path reuse,solving the bias problem in bidi-rectional path reuse,analyzing and alleviating the covariance problem in unidirectional path reuse,and utilizing the sample coherence in image space to develop a high performance fur rendering algorithm.Specifically,this thesis consists of the following content:(1)For bidirectional path reuse,we propose a novel photon gathering method as a replace-ment of the density estimation to efficiently achieve unbiased rendering with photon mapping.Specifically,instead of aggregating the gathered photons,we process each photon individually and connect the corresponding light sub-path with the eye sub-path that generates the gather point.This connection can be interpreted as a Russian roulette event decided by the gather point,and its probability is an integral.As this probability is in the denominator of the final estimate,one can not efficiently estimate it without introducing bias into the algorithm.Therefore,we introduce an unbiased algorithm to estimate the reciprocal of the roulette probability using a series of Bernoulli trials.The end result is a self-contained unbiased sampling technique,which can be incorporated into a wide variety of Monte Carlo light transport algorithms.(2)For unidirectional path reuse,we study the impact of path reuse covariance in the final estimation variance.The estimation variance consists of two parts,the variance term and the covariance term of the reconstruction samples.Our study shows that increasing the reconstruction rate can only reduce the variance term while the covariance term remains almost unchanged.We also find that the covariance term actually represents the extent of reconstruction-time reuse of indirect light samples.The more reconstruction samples reuse the same indirect light sample,the more significant the covariance term becomes.Based on this variance analysis,we propose a scheme to adaptively generate indirect light samples to reduce the covariance term.We applied this method to indirect light field reconstruction and axis-aligned image space filtering,and reduced the artifacts in rendering results significantly.(3)For image space shading reuse,we present a cone-based ray tracing algorithm for high-quality rendering of furry objects.By aggregating many sampling rays in a pixel as a single cone,we can actually reuse the sampling results within the cone,and significantly reduce the high su-persampling rate required by the thin geometry of fur fibers.This allows us to we reduce both the sampling overhead and the fur compositing cost.The result is a highly efficient ray tracing algo-rithm for furry objects which is able to achieve a rendering quality comparable to state-of-the-art alternatives,while significantly reducing the rendering time.
Keywords/Search Tags:Photorealistic rendering, Monte Carlo ray tracing, path reuse, photon mapping, indirect light field reconstruction, fur rendering
Related items